We’re looking for hackers, builders, researchers and engineers

We build agents, study what makes them work, and ship what we discover to thousands of users. If you want to work on problems like that, write to us.

What we work on
Shipped to

VisitBlekinge, Power Circle, Northbike, 360 Träningscenter, Kaffebrus, Rally Sweden, CyberPhoto, Citykompaniet, Scandinavian Photo, Swoosh, Billig Teknik.

Fields we work in

Psychology

Why people trust a system that is sometimes wrong, and what changes when they can watch it reason.

Behavioural science

How incentives shape what an agent does, and what people start doing once an agent is in the loop.

Neural networks

Architecture and training dynamics, and which representations actually form inside a model.

Game theory

Multi-agent settings where cooperation is not assumed and defection is cheap.

Agents

Autonomy with a mandate and a stop condition. What holds when the environment pushes back.

Machine learning

Evaluation over intuition. Measuring capability honestly, including where it quietly fails.

Bots

Systems that act continuously without waiting for a human turn. Latency, drift and containment.

AI compute

What is reachable per unit of compute, and where the real cost sits inside an agent loop.

First principles tell us where to start. Data tells us whether we were right.

We break a problem down to what we can actually verify, build up from there, and put the result in front of real users. Then we measure, and go again.

Questions we ask

If you have an opinion on any of these, that’s reason enough to write to us.

How much of what a person wants can be known before they say it, and how little data does that take?

Can you measure whether an agent understood a task separately from whether it completed one?

What is the smallest amount of human review that still keeps an autonomous loop safe?

How many mistakes can an agent make before people stop trusting it, and can that trust be rebuilt?

When an agent is paid by the seller and trusted by the buyer, what should it do when the right answer is "don’t buy this"?

Can you tell persuasion from manipulation by looking only at the logs?

When two agents negotiate on behalf of two humans, whose preferences survive the exchange?

How do you prove an agent caused an outcome, rather than just being there when it happened?

Is chat the final interface for intelligence, or just the first one that worked?

What is the real limiting factor on a useful agent today: intelligence, context, or permission?

If the underlying model gets ten times better next year, which of these problems disappear and which get worse?

Skip the CV

Send us something you’ve built, a problem you cracked, or an argument for why one of the questions above is wrong. A half-finished repository tells us more than a polished cover letter ever could.

A founder reads every message, and you’ll hear back within a week either way. No recruiters. No five-round process.

We pay for what you make happen, not for the hours you log.

or just write to samir@nordenai.se

Probably you

  • You have an unfinished project you still think about.
  • You would rather be measured than believed.
  • You read well outside your own field.
  • You have changed your mind about something important.

Probably not

  • You want a defined role with a defined scope.
  • You need the problem specified before you start.